Lead AI Engineer - \"Local-first LLMs, real production ownership\" - Pune

CareerXperts Consulting; Pune District; On-site; INR 1,800,000 - 3,000,000; Full time

Job summary

CareerXperts Consulting in Pune, India is seeking a player-coach to lead an ML/agentic AI layer. You will set technical direction for a team of AI/ML engineers while staying hands-on with the hardest problems, spanning classical ML and LLM-driven agentic reasoning.

You’ll own the ML layer, agentic workflows, and model deployment decisions across private cloud and sovereign customers, with strong emphasis on Python, PyTorch, and cloud-based MLOps. Prior experience with vLLM/Ollama is a plus.

Qualifications

• 8+ years in ML/applied AI/data science.
• 2+ years leading or tech-leading an AI/ML team, still hands-on.
• Proven ownership of ML + agentic AI in production.
• Strong RAG, agent orchestration, prompt engineering, fine-tuning.
• Hands-on with vLLM/TGI/Ollama — quantization, GPU sizing, throughput/cost trade-offs.
• Python (PyTorch/TensorFlow/scikit-learn), SQL, MLOps on AWS/GCP/Azure with Docker/K8s.
• Bonus: UEBA, graph analytics, MITRE ATT&CK/ATLAS familiarity, publications or open-source contributions.

Responsibilities

• Own the ML layer — alert classification, case prioritization, behavioral baselining, feature engineering over security telemetry.
• Lead agentic workflows — retrieval-grounded investigation, tool use, structured output.
• Make model decisions — where to run, private-cloud & sovereign deployments, balance inference cost/latency/quality at scale.

Skills

Python

SQL

MLOps

Docker/Kubernetes

PyTorch/TensorFlow/Scikit-learn

RAG

Prompt engineering

Leadership

Agentic AI

Tools

vLLM

Ollama

GPU sizing

quantization

Job description

We're building the ML and agentic AI layer that decides what deserves an analyst's attention — and we need a player-coach to lead it.

This isn't a management-only seat. You'll set technical direction for a team of AI/ML engineers while staying hands-on with the hardest problems yourself — spanning classical ML (classification, ranking, clustering, anomaly detection) and LLM-driven agentic reasoning on top of it.

You’ll own:

• The ML layer — alert classification, case prioritization, behavioral baselining, feature engineering over security telemetry
• Agentic workflows — retrieval-grounded investigation, tool use, structured output
• Model decisions — what runs where, local/self-hosted deployment for private-cloud & sovereign customers, inference cost/latency/quality at scale

What you bring:

• 8+ years in ML/applied AI/data science, 4+ owning production systems
• 2+ years leading or tech-leading an AI/ML team, still hands-on
• Proven ownership of ML + agentic AI in production (not just prototypes)
• Strong RAG, agent orchestration, prompt engineering, fine-tuning
• Hands-on with vLLM/TGI/Ollama — quantization, GPU sizing, throughput/cost trade-offs
• Python (PyTorch/TensorFlow/scikit-learn), SQL, MLOps on AWS/GCP/Azure with Docker/K8s
• Bonus: UEBA, graph analytics, MITRE ATT&CK/ATLAS familiarity, publications or open-source contributions.

If you've shipped both the "narrow the signal" and "reason over it" halves of this problem — and want real ownership over which — let's talk.